基于改进的PSO算法的神经网络相关性剪枝优化*

针对传统的神经网络训练算法收敛速度慢、易陷入局部最优的问题，提出了一种基于改进的分期变异微粒群优化算法（SMPSO）的神经网络相关性剪枝优化方法。SMPSO在初期使适应度过低的微粒发生变异，在后期使停滞代数过高的个体极值和全局极值发生变异,后将SMPSO用于优化神经网络相关性剪枝算法。实验结果表明，该方法与采用BP算法及标准PSO算法进行相关性剪枝相比，在训练收敛速度、剪枝效率及分类正确率三方面都有较大提高。

Neural network correlation pruning optimization based on improved PSO algorithm

The traditional neural network training algorithm converges slowly and is easy to fall into local optimum. In response to these shortcomings, this paper proposed a neural network correlation pruning method optimized with improved staging mutation particle swarm optimization algorithm (SMPSO). SMPSO mutate particles that had too low fitness at early stage and mutate individual extreme and global extreme that stagnate in excessive iteration latterly. Then used SMPSO to optimize neural network correlation pruning algorithm.The experiment results show that neural network correlation pruning method optimized by SMPSO is more efficient than that optimized by BP and standard PSO. It has greater improvement in the convergence velocity of training, the efficiency of pruning and the accuracy of classification.